MPS Powered by Eureka

Submission Tilte

Federated Learning for Secure Medical Imaging Data Sharing in Healthcare Networks.

Submission Abstract:

Decentralised machine learning model training is known as federated learning. Instead of sharing the original data, federated learning shares model changes, such as gradient information, which inherently safeguards the data kept on each device. But since model updates are dependent on the original data, they may expose private information. For healthcare datasets, logistical, ethical, or privacy issues frequently make it hard to aggregate data samples from several locations. Without requiring data pooling, federated learning enables the use of potent machine learning algorithms. Managing related workflows and ensuring the interoperability of systems used for medical image production, storage, sharing, display, transmission, querying, processing, retrieval, and printing are its goals. Data transfer from client devices to global servers is not necessary. To improve data privacy, the model is trained locally using the raw data from edge devices. A unique type of distributed learning architecture called federated learning permits several users to take part in model training without compromising their privacy; yet, this paradigm is still susceptible to security and privacy risks from different attackers. By training algorithms jointly without sharing the data itself, federated learning aims to solve the issues of data governance and privacy. The common practice of centralising data from several centres nowadays comes at the expense of serious issues with data security and patient privacy. A global model is initialised on a central server as the first step in the Federated Learning process. After that, local devices like smartphones and edge devices receive this global model. Locally updated model parameters are the outcome of each device autonomously training the model using its own local data. Distributed clients can work together on training thanks to federated learning, which also protects client privacy by storing data locally. However, because data, models, and devices are heterogeneous, the final global model might need to perform better for jobs on each client. Federated learning speeds up AI and IoT deployment models and increases operational flexibility by enabling multi-location dataset storage and simultaneous training of models on each piece of data. Medical imaging, sometimes referred to as radiography, is the branch of medicine where doctors make different images of bodily components for diagnostic or therapeutic purposes. Non-invasive tests used in medical imaging techniques enable physicians to diagnose illnesses and injuries without causing undue discomfort. Medical imaging aims to diagnose and treat illnesses by exposing internal structures that are concealed by the skin and bones. In order to detect anomalies, medical imaging also creates a database of normal anatomy and physiology. A distributed machine learning framework termed federated learning (FL) is becoming more popular as demands for protecting the privacy of health data rise. The server manages the learning process, combines client updates to enhance the global model, and then sends the revised model back to the clients. In a federated learning framework, different clients use their local data to participate in the training process in a distributed fashion. To be precise, each client trains a model on its own using their dataset and then distributes the model parameters to other participants. A variety of fields and viewpoints are welcome to contribute, including but not limited to: Federated Learning for Secure Medical Imaging Data Sharing in Healthcare Networks.